- Todo List/Requests for Startups (ongoing post)
Just now·3 min read
I want to see each of these come up in tomorrow’s TechCrunch so bad!
Credit: BigStock Image via GeekWire
Pagerank for “Expert Opinion” — Brian Chesky said in this interview with Reid Hoffman, that the key to being an infinite learner is finding the source of truth on any question asap. Doing self-study of biology, I can attest that the importance of a focused question and finding the right source is extremely important.
Not only that, but with all the misinformation, feeds optimized for advertising and virality, and COVID breeding distrust of institutions it’s important to have transparent evaluations of internet gurus, lest they lead us even further astray.
An MVP here might be a twitter bot that can record a % trust in a person and the topic/question for which that trust is ascribed. Users could then search by other experts, their friends, city, etc to see who is trusted when they see responses on their feed. Maybe you could even mechanical turk check for correctness longitudinally and update trust through time (perhaps start in a specific vertical/question).
Another possibility might be a bot which can take a hashtag/question and cluster over an embedding for the tweets in the space. Then display some visual of the groups to the user and color-code responses accordingly. Each of these first steps clearly misses a lot of problems.
I started work on this for a group project last quarter (repo here), it’s focused on Google News results, and there is a chrome plugin by the same name — Pulse — but under my partner’s name Dean Stratakos. Check the * at the end for more info on my current todos to improve the project.
Make running iPOP on yourself an accessible, everyday reality — tracking the development of diseases you’re at risk for and being able to debug the transcription, metabolic problems, etc is absolutely bonkers and would certainly change the role of traditional healthcare, putting the power to pursue preventative measures directly in the hands of consumers. Furthermore, if it’s possible to combine this with projects that are finding ways to do anonymized, open-source sharing of medical data across institutions, this would probably spur a flurry of medical advances from diagnosis to therapeutics.
How does search need to change in a world that is wearables first (health trackers, AR glasses, voice assistants), when the internet we are computing on is the whole world and not just text?
Make laptops a subscription service with cloud instances streamed to a thin client, same with phones. Also, please make them iPad Pro style so there is laptop productivity and touch screen creative options.
*We focused on news, but I think our positive/negative/neutral restriction is too severe (due to using an existing dataset for the class) and doesn’t really address the problem we wanted to address, which is mapping the clusters of opinions on a subject and showing how your feed is being biased. It is also not specific to the question, which could flip the desired behavior with a simple not. That could be solved with just swapping the model for something with masking and pulling the subject from the sentence with typical NLP parsing models. Last component missing here imo is being able to pull more of the article text for the prediction. A solution here might be subscribing to major outlet RSS feeds and cacheing the predictions, then later regurgitating them upon a user’s request.
Inspiration for this post came from Will Robbins at Contrary Capital
- Date of publication:
- Tue, 05/04/2021 - 15:21
Click on the link - it will be copied to clipboard